Re: [PATCH v1 RFC 1/2] spi: introduce fallback to pio

[Vinod Koul <vkoul@xxxxxxxxxx>]

Hi Robin,

On 15-06-20, 08:59, Robin Gong wrote:
> On 2020/06/15 15:20 Vinod Koul <vkoul@xxxxxxxxxx> wrote:

> > > Yes, but both assume spi controller driver could detect such dma
> > > failure before dmaengine_prep_*(). Let's wait Vinod's comment for that
> > > if dmaengine_slave_config could keep direction.
> > 
> > The direction is already in the prep_ call, so sending in dmaengine_slave_config
> > is not required, pls pass it in the prep_ call
> Hi Vinod,
> 	Is there any way to let the device driver to know dma controller is ready
> (in sdma case is sdma firmware loaded or not)before prep_call? Hence, spi core
> could map dma buffer or not. Prep_call is too late for spi core since the buffers
> have been already mapped. 

Can you use .device_alloc_chan_resources for that? This is where all
the resource allocation for a channel should happen...

> 	From my view, seems dmaengine_slave_config is the only one...Further,
> sdma need direction in dmaengine_slave_config phase, because currently
> what's the tx/rx script used on sdma channel is decided not only peripheral_type
> but also direction. For example, spi tx dma is running ram script to workaround
> ecspi ERR009165 while rx dma is running rom script, so only spi tx dma channel
> depends on sdma firmware loaded(now that could be detect by ' load_address
> < 0' in sdma_load_context() and prep_ call finally).
>    I knew direction is deprecated in dmaengine_slave_config, but that's really
> very useful for sdma to check if firmware loaded and spi core could get it earlier
> before prep_call(fallback to PIO if dma is not ready).

I think that is wrong expectation, dmaengine_slave_config should pass
the slave_config to driver and nothing else. The relevant action should
be taken in respective prep_ calls here, so that should be fixed as well

-- 
~Vinod